Political Shirking and the Last Term Problem: Evidence for a Party-Administered Pension System

Political Shirking and the Last Term Problem: Evidence for a Party-Administered PensionSystemAuthor(s): John CareySource: Public Choice, Vol. 81, No. 1/2 (Oct., 1994), pp. 1-22Published by: SpringerStable URL: http://www.jstor.org/stable/30027102 .

Accessed: 14/06/2014 03:14

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

Springer is collaborating with JSTOR to digitize, preserve and extend access to Public Choice.

http://www.jstor.org

This content downloaded from 62.122.79.69 on Sat, 14 Jun 2014 03:14:22 AMAll use subject to JSTOR Terms and Conditions

http://www.jstor.org/action/showPublisher?publisherCode=springer

http://www.jstor.org/stable/30027102?origin=JSTOR-pdf

http://www.jstor.org/page/info/about/policies/terms.jsp


Public Choice 81: 1-22, 1994. @ 1994 Kluwer Academic Publishers. Printed in the Netherlands.

Political shirking and the last term problem: Evidence for a party-administered pension system*

JOHN CAREY Department of Political Science, University of California, San Diego, CA 92037

Accepted 7 October 1992

Abstract. Studies of political shirking have disagreed both over whether the voting behavior of Members of Congress changes in their last term, and over the manner in which last term shirking can be controlled: through electoral sorting, or through a pension system. This paper presents evidence that Members of Congress who leave the House to run for statewide office do alter their voting behavior between the two sessions of their last House term, and that this change includes an ideological shift toward their state party delegations. The results suggest that a party-driven pension system influences the voting of House members who aspire to higher office, but that the pension system is not sufficient to control the last term shirking likely to occur if term limitations were imposed on House members.

1. Introduction

One of the fundamental problems of representative government is ensuring that elected officials accurately represent the interests of their constituents. The most straightforward solution to the problem is to rely on the desire for reelection to motivate good behavior. Provided that representatives value holding office, and can seek reelection, they will act as good agents for their constituents, or else run the risk of being thrown out. The question that follows, of course, is how to ensure that a representative serving her last term, and so not subject to electoral control, will not shirk the responsibility to serve her constituents. This question becomes all the more pressing in the context of the current debate over whether legislators at both the state and national level should be subject to term limits. In this paper, I consider two general types of answers to the last term question - one that relies on electoral markets as enforcers, and one that relies on political parties - and I present some new data on the effects of parties.

The paper will proceed as follows. First, I examine how shirking has been

* The author wishes to thank Jonathan Katz, John Lott, Jr., Keith Poole, Michael O'Hagan, and especially Gary Cox, for assistance and constructive criticism in preparing this paper. All short- comings were provided by the author.



2

defined and the methods used to detect it. I also address the question of whether such shirking necessarily implies a loss of voter control over politicians. Next, I consider the relation between shirking and a politician's last term, and describe the two principal methods by which the last term shirking can be controlled in theory: sorting and pensions. I review the evidence provided by the public choice literature as to whether and under what circumstances significant shirking exists. Then I conduct a series of tests that demonstrate that members of congress who aspire to statewide office do alter their voting behavior during their last term in the House of Representatives, and that political parties influence the voting of these aspirants, by controlling their opportunities to advance to higher office. Finally, I draw some conclusions about the effectiveness of such party-administered pension systems in providing for electoral control of politicians, particularly in conjunction with the prospect of constitutional term limitations for members of Congress (MCs).

2. Shirking

2.1. How to measure

In recent years many studies have confronted the methodological problem of determining whether shirking takes place, presenting evidence of who shirks, how much, and when.1 Much of this work has been devoted to detecting and measuring shirking detrimental to constituent interests; some has addressed the last term problem as well. The fundamental question inevitably is what is the responsibility that is or is not "shirked"?2 Most analyses of shirking have sought to identify a battery of constituent interests based on economic and demographic data from a representative's district, and then to test how accurately the representative's votes reflect these interests (Kau and Rubin, 1979; Peltzman, 1984; Lott, 1987; Higgs, 1989; and Bender, 1991). The extent to which the representative's votes do not reflect constituent interests is some- times taken to indicate the amount of personal ideology she is "consuming" on the job, and such consumption is considered shirking. Others have suggested that shirking can be non-ideological and oriented toward maximizing the value of holding office or the probability of reelection. I agree with Bender (1991) that it is unimportant to constituents whether their interests are being ignored for ideological or other reasons. However, it is important to determine whether responsiveness to some other set of interests replaces responsiveness to constituent interests among last term politicians.

In the standard methodology, a representative's vote on a given issue, or an index of her entire voting record, is taken to be a function of the interests of her constituency and her proclivity to shirk. Voting as the dependent variable



3

is regressed on a vector of demographic and economic variables representing constituent interests, as well as on a variable representing ideological consumption (or shirking). The more closely voting corresponds to the values of the constituent interest variables, the lower the residual that is attributed to shirking. Higgs (1989) examined the voting of senators from the same state, assuming that when representatives from the same constituency vote differently, at least one must be shirking constituent interests.3 However, Lott (1990b) and Lott and Davis (1992) counter that senators from the same state can represent different constituencies, and so vote differently without shirking. This fits with Fen- no's (1978) well-known study of the ways in which representatives build personal support coalitions in their districts. MacArthur (1990) and Lott (1990a) have measured attendance at roll call votes to determine whether some politicians shirk by consuming leisure, as opposed to shirking by voting against constituent interests.

One of the primary differences among the shirking studies is whether cross- sectional or time series data are used. Where data are cross-sectional, absolute levels of shirking can be measured for each representative, and relative levels can be measured among representatives for that one cohort at that time. For example, one could determine whether representatives in their last term shirk more than others. Only by using time series data, however, can one determine whether specific representatives shirk more in their last term than they had previously. The use of time series data, then, is essential to testing for whether the behavior of a given representative is likely to change in the last term. And this question must be at the center of any study of the effects of term limits.

2.2. Does shirking imply a loss of electoral control?

A normative question this literature has raised is whether shirking necessarily constitutes a loss of control over politicians by the electorate. Lott (1990b) has suggested that individual shirking will not necessarily matter to policy output if political markets are self-correcting. Voters may elect candidates who balance perceived discrepancies between voter preferences and the policy outputs of the government as a whole. Thus, if policy is too liberal for voters, they compensate by electing a representative more conservative than their policy ideal. If political markets were entirely self-correcting, however, voters would act without regard to their attitudes toward specific candidates, and only with regard to their attitude toward overall policy outputs. But we know that representatives consider the implications of their own specific votes on prospects for reelection, and that voting records are often used effectively by challengers to office. Self-correcting markets, then, cannot render irrelevant the issue of shirking for constituent-representative relationships.



4

Other work suggests that a representative's votes may not correspond with the interests of her district for a number of reasons. Kau and Rubin (1979) find evidence of stable coalitions in the Senate that logroll in order to pass packages of particularistic legislation, each element of which may be in the interests of only a few coalition members. Standard methods of detecting shirking do not allow for vote trading, and so would interpret logrolling as shirking. Both Peltzman (1984) and Fenno (1978) argue that representatives do not respond to their entire district, but rather to the coalition of voters that support them and the monied interests that finance their campaigns. Again, the argument is that a vector of variables reflecting the interests of an entire district will not accurately show a politician's responsiveness to those who elected her. More- over, it is possible that representatives maintain ideological consistency across all votes, even at the occasional expense of district interests, in order to send clear signals to the electorate what pattern of representation can be expected. Such clear signaling can be an investment in electoral capital if voters value predictability as well as interest voting (Dougan and Munger, 1989; Peltzman, 1984; Lott, 1987).

All these explanations - logrolling, coalition-building, and signaling - explain patterns of legislative behavior observationally equivalent to shirking as efforts to garner electoral support. Yet if the last term problem generates shirking, it is precisely because representatives are no longer concerned with garner- ing electoral support. None of these explanations, then, can account for any shirking that takes place due to the last term problem, nor do any of them imply that last term shirking is not relevant.

3. The last term problem

The last term problem can be described as the destabilization of a cooperative equilbrium between representative and constituency.4 Cooperation for representatives entails acting as good agents of constituency interests; and for voters it entails rewarding good agents with reelection. As long as both sides view the relationship as indefinite - or at least without an identifiable last period - cooperation is possible, and electoral control can function. Once the relationship is viewed as finite by either party, however, rational players will defect, and the game will unravel. The two principal solutions to the last term problem are (1) to create some sort of pension system for representatives who have performed adequately all the way through the last period, or (2) to rely on the electoral market to sort potential shirkers out of office.



5

3.1. Pensions

The logic of pensions developed by Becker and Stigler (1974) is generalizable to all principal-agent relationships, including that between constituents and their elected representatives. An employee5 who approaches her final pay period will be increasingly tempted to shirk, since the benefits will increasingly out- weigh the wages foregone if shirking is detected. The solution is a compensation system in which employees who might gain by shirking are rewarded for good behavior with a bonus that is positively related to the potential gains from malfeasance, and inversely proportional to the risk of detection. Thus, as the potential gains from shirking increase, so does the bonus needed to prevent shirking. But as the odds of detecting malfeasance grow, the size of the bonus can be reduced. The bonus is, of course, paid only on satisfactory completion of the last pay period, and is forfeited otherwise. Thus, the prospect of losing the pension is increasingly frightening as one approaches retirement.

Arguing along essentially the same lines, Barro (1973) suggested that political parties might serve as enforcers of good behavior among last term representatives by administering pension systems. Below, I shall examine whether there is evidence that political parties play this role, insofar as they can deliver or withhold the "pension" of nominations and electoral support for House members seeking for higher office.6

3.2. Sorting

Lott and Reed (1989) argue that the last term problem is already solved, not by an enforcer, but prior to the vaunted last term itself, by market forces. The logic is that representatives who are ideologically compatible with their constituencies naturally perform so as to maximize the probability of reelection. The opportunity costs of good agency are lower for such representatives than for those who are ideologically incompatible with their constituents. Competition in producing good performance, and credible signals that good performance will continue, yields too low a return to the incompatibles, so they shirk more. Thus potential shirkers are sorted out of office by the electoral market before voluntary retirement is an option. Surviving representatives are exceptionally compatible with their constituents, and so there is little agency loss even in the last term (Lott and Reed, 1989).

Two additional points should be made regarding the sorting argument. First, sorting need not be immediately thorough to overcome the last term problem. Where representatives generally serve multiple terms, sorting can diminish last term shirking as long as it removes from office potential shirkers over a series of reelection bids. To check shirking in the context of compulsory term limits,



6

however, sorting must work quickly. Second, even if sorting works, those representatives who make it to retirement will not necessarily be naturally perfect agents. They will merely be those for whom the opportunity costs of good agency were lowest. Their last term incentive will still be to shirk. So we can imagine a combination of the sorting and shirking effects, in which retiring politicians shirk more than they did in previous terms, but less than do the cross-section of their peers.

4. The existing evidence

While studies of the last term problem have yielded differing results, the work as a whole suggests two broad conclusions. First, the sorting argument is borne out by cross-sectional analyses of MCs showing that those who retire tend not to shirk in their last term any more than do non-retirers (Lott, 1987; Lott and Reed, 1989; Dougan and Munger, 1989).7 Second, Zupan's (1990) time series analysis indicates that retiring congressmembers shirk more as they approach retirement than they did previously.8

There are a number of studies that offer various evidence that last term shirking should cause negligible slack in the electorate's control over politicians. Lott (1987) and Lott and Reed (1989) provide evidence demonstrating that retiring MCs as a group are better agents of their districts than non- retirers, shirking less in their last term than do cross-sections of their peers. Dougan and Munger (1989) present time series evidence that senators do not shirk more when more years remain until their next reelection bid. This suggests that shirking is unrelated to electoral maneuvering, provided we assume that politicians discount the future somewhat. VanBeek (1990) detects no increase in shirking in the months betwee MCs' decision to retire and the end of their terms. Moreover, Lott and Davis (1992) provide a time series test of shirking among senators to determine whether shirking increases with tenure, with years until the next election, or in the last term. They find no consistent correla- tion between shirking and any variable except whether a senator is subsequently defeated at the polls. Their evidence suggests that electoral markets are quite effective in sorting shirkiers out of office.

Contending against these arguments are a number of studies suggesting for various reasons that last term shirking does cause slack in the constituent- representative relationship. Kau and Rubin (1979) find that MCs consume ideology (shirk) against both vectors of constituent interests and logrolling coalitions. Bender's (1991) study of voting on campaign finance reform laws shows that MCs first and foremost maximize the value of holding office, and honor ideological obligations only when the opportunity costs are not prohibi- tive.9 Davis and Porter (1989) find that shirking against their states' interests



7

increases with age once senators pass a threshold age of 53 years. These results suggest that the last term temptation to shirk applies more generally as the probability that a politician is in her last term increases. Kalt and Zupan (1990) find that senators shirk more the more political capital (committee chairman- ships, large margins of past electoral victories) they possess, the longer until they face reelection, and in their last terms before retirement. However, the Kalt and Zupan methodology is subject to a simultaneity bias, since it is not clear that causality runs in only one direction between their dependent and independent variables - shirking and levels of political capital.

In an effort to correct for this problem, Zupan (1990) examines changes in the voting behavior of House members between the sessions of the same Con- gress. He distinguishes among those members who run for reelection, those who decide to retire prior to the last term, those who decide to retire during the last term, and those who leave the House to run for other offices. Zupan finds that retiring members shirk significantly more in the second session than the first, and this pattern is particularly pronounced for late deciders and aspirants to other offices. Despite this pattern, however, Zupan finds that absolute levels of shirking are no greater among retirers and aspirants than among the entire cross-section of members. This supports the argument that the sorting effect removes rampant shirkers before retirement. Third, aspirants as a group exhibit low absolute levels of shirking despite the fact that they shirk more in the second session than in the first. Thus it seems that exceptionally good agents are rewarded with the opportunity to run for higher office.

Much more attention has focused on determining whether shirking exists than on whether political pensions do. Nevertheless, two studies have included limited tests of Barro's (1973) suggestion that parties might prevent shirking by controlling the future careers of officeholders. Peltzman (1984) holds that the key determinants of voting are the nature of a senator's support coalition and her interest group contributors. He tests whether national party organiza- tions might affect voting behavior as any other interest group does - according to the proportion of campaign funding a senator receives from the party. Peltz- man's results are negative, and suggest that parties may actually fund shirkers at a greater rate than non-shirkers. His test, however, compares funding from national party organization with the level of shirking against state constituencies. There is no reason to believe that the interests of these two entities coincide.

Lott (1990a) and McArthur (1990) have shown that, regardless of how they vote, retiring congressmembers shirk by consuming leisure on the job, with levels of absenteeism almost twice as high as for non-retirers. Lott goes on to determine, however, that among members who have children running for elective office, the increase in absenteeism is only half so great. And for members pur- suing post-congressional careers in lobbying or other government work, who



8

also have children in politics, there is no increase in absenteeism. Lott's results suggest that some sort of pension system based on future career opportunities exists to deter at least one type of shirking among certain retirers.

To this point, it is evident that the logic of sorting does not preclude a rise in shirking in the last term among heretofore good agents. Zupan's (1990) results support this interpretation, particularly for aspirants. The combined evidence from all the studies reviewed tell the story that sorting removes potentially big shirkers, and that the last term effect remains for retirers. But the question of pensions remains open. If aspirants increase the level of shirking against their districts in their last year, how could a pension system be working? The answer, of course, is that if pensions are administered by parties, then to examine patterns of shirking against the old district will tell us nothing about the quality of enforcement. We need to know whether aspirants shirk against their party, and ultimately against their state. If political parties do administer pension systems, then the last term effect of shirking against current constituents should be accompanied by increased responsiveness to the interests of state parties. While Barro (1973) and Becker and Stigler (1974) almost twenty years ago suggested that parties may act as enforcers through control over post- retirement career opportunities, none of the studies of last term behavior tests for such a phenomenon.

5. A test for state and party pensions

The test conducted here determines whether aspirants move systematically toward either their state or their state party interests at the expense of their current constituencies during the second session of their last House term. Such movement indicates a trade-off for responsiveness to those who control access to higher office against responsiveness to current district interests.10 The first step is to demonstrate that aspirants alter their voting behavior more between the two sessions of their last Congress than do non-aspirants. I also offer evidence that the shift in voting behavior among aspirants is greater the more ideologically extreme is the MC. The next step is to demonstrate that this increased movement between sessions can be accounted for as movement toward the aspirant's respective delegations (the state delegation, and the state party delegation). Again, the effect is greater as the aspirant's ideological extre- mism increases. Finally, I present evidence that it is movement toward the state party delegation that distinguishes aspirant's voting behavior from that of non- aspirants.

I use NOMINATE scores as indicators of individual voting patterns, and mean NOMINATE scores of state, and state party, delegations as indicators of state and party interests."11 NOMINATE (NOMINAl Three-step Estima-



9

tions, developed by Poole and Rosenthal, 1985) scores are derived from all non-unanimous roll call votes taken during a session of Congress, and register each MC's location on a single liberal-conservative dimension. The scores are normally distributed and centered on 0. Those used here have a standard deviation of .59, and at the extremes range from - 2.23 on the liberal end to 1.96 on the conservative end.

The first test determines that aspirants do in fact alter their voting behavior more between the first and second session of their last House Congress than do non-aspirants. The average magnitude of the difference between first and second session NOMINATE scores for non-aspirants is .096. But among aspirants the average magnitude of the difference is .023 points higher, this difference being discernible at conventional levels of significance.12

Another way of demonstrating this point is to show that the first session NOMINATE score predicts the second session NOMINATE score less effectively for aspirants than for non-aspirants. The relevant OLS regression is:

NOMINATE2= - .001(CONSTANT) - .004(ASPIRANT) (1) (- .276) (- .382) + .980(NOMINATE1) - .053(NOMINATE1 *ASPIRANT)

(286) (-3.01) Adjusted R2 = .952,

where

NOMINATE2 is the second session NOMINATE score;

CONSTANT is set equal to 1;

ASPIRANT is a dummy variable scored 1 if an MC leaves the House after that term to run for statewide office, and 0 otherwise;

NOMINATE1 is the first session NOMINATE score; and

NOMINATE1*ASPIRANT is an interactive term designed to pick up the difference between the effectiveness of NOMINATE1 as a predictor of NOMINATE2 for aspirants as opposed to non-aspirants.

The constant term determines whether there is any systematic difference between NOMINATE scores between sessions for non-aspirants - that is, whether such MCs are consistently more liberal or conservative in one rather than the other. The ASPIRANT dummy determines the same thing for aspirants only. The negligible and non-significant coefficients of these terms indi-



10

cate that there is no such systematic movement for either set of MCs. The coefficient of NOMINATE1 and the t-score show that second session voting behavior tends to follow first session voting closely. And the coefficient and t-score for the NOMINATE1*ASPIRANT variable show that second session voting follows first session voting significantly less closely for aspirants than for non-aspirants. The .980 slope coefficient for aspirants falls to .927 among non-aspirants, and the groups are statistically discernible at conventional significance levels.13 This confirms the result of equation (1), that aspirants alter their voting behavior more than non-aspirants between the sessions of their last Congress. Second, the equation suggests that the more NOMINATE scores differ from 0, the more voting behavior changes between sessions. The interactive variable is constructed such that its absolute value grows as the NOMINATE score increases in magnitude - that is, as the MC becomes more ideologically extreme. And as the value of the interactive term grows, the same coefficient predicts a greater change in voting behavior for the aspiring MC.

At first glance, this may appear puzzling. Why should changes in voting behavior among aspirants grow as the magnitude of the NOMINATE score grows? There are two possible reasons for this - one of which applies to all MCs, and the other only to aspirants. First, the estimation of ideological position with NOMINATE is more difficult for more ideologically extreme, or pure, legislators than for moderates (Poole and Rosenthal, 1985: 364-365). Given the estimation problems, we might expect more variance in scores between sessions among more extreme MCs, whether they are aspirants or not. This is in fact the case. Nevertheless, it is also reasonable to expect that extremists who are also aspirants will shift their voting even more dramatically than non-aspiring extremists, if they are attempting to court a new, statewide constituency; and more than moderate aspirants, given that statewide constituencies are likely to be more heterogenous, and so less ideologically pure, than House districts.

The following equation tests these hypotheses:

DIFFERENCE = .074(CONSTANT) - .042(ASPIRANT) (2) (30.2) (- 3.13)

+ .044(MAGNITUDE NOMINATE1) (10.8)

+.118(MAGNITUDE NOMINATE 1 *ASPIRANT) (5.71) Adjusted R2 = .042

where

DIFFERENCE - is the absolute value of the difference between first and second session NOMINATE scores for an MC;



11

CONSTANT - is set equal to 1;

ASPIRANT - is the dummy variable used above;

MAGNITUDE NOMINATE1 - is the absolute value of the first session NOMINATE score; and

MAGNITUDE NOMINATE1*ASPIRANT - is an interactive term designed to pick up any difference between the effect of ideological extre- mism on DIFFERENCE for aspirants versus for non-aspirants.

These results refine our picture of aspirant voting patterns considerably. In the first place, the positive coefficient for MAGNITUDE NOMINATE1 demonstrates that, among non-aspirants, those who are more ideologically extreme also show greater differences between their scores from session to session. This could be a result of the estimation difficulties among extreme legislators. But the coefficient for the interactive variable indicates that this effect is 4 times greater among aspirants than among non-aspirants, and the t-score shows that the difference between the groups is strongly significant.14

Moreover, the slope coefficients and the intercept values together tell a more precise story. The negative coefficient on the ASPIRANT dummy indicates that, among ideologically moderate MCs, aspirants actually vary less between sessions than do non-aspirants. Among MCs for whom the absolute value of NOMINATE1 is less than .36, aspirants show less volatility than non- aspirants.15 As MCs become more ideologically extreme, however, aspirants move more between sessions. The dramatic movement among more extreme aspirants is sufficient to generate the initial results that, overall, aspirants are more erratic in their voting patterns than non-aspirants.

The results so far indicate that:

(1) Overall, aspirants to statewide office change their voting behavior between sessions of their last term in the House more than do non-aspirants; but

(2) The voting pattern of moderate aspirants are more stable than those of non- aspirants. Those who change the most between sessions are those aspirants who are originally the most ideologically extreme.

These points suggest a scenario similar to that proposed by James Madison (1961) in Federalist #10, wherein larger and more heterogenous constituencies demand more moderation on the part of their representatives than smaller ones. If this "heterogeneity breeds moderation" hypothesis is apt, then statewide constituencies should be more moderate on the whole than House

districts; and it follows that moderate aspirants should not have to change their



12

voting behavior as much as extremists in order to appeal to the statewide constituency.

This generates the question of whether the changes in voting behavior among aspirants detected above can be explained as movement toward the new statewide constituencies the aspirants seek to attract. If so, then the dramatic

changes in voting behavior among extreme aspirants detected in equation (2) should translate into movement toward the state delegation. The following equation tests this hypothesis:

MOVEMENT TOWARD STATE = -.047(CONSTANT) -.081 (ASPIRANT) (3) (-6.58) (2.06) + .082(MAGNITUDE NOMINATE1) (6.80) + .19(MAGNITUDE NOMINATE1 *ASPIRANT) (3.08)

Adjusted R2 = .016

where

MOVEMENT TOWARD STATE - is the magnitude of the difference between an MC and the purified mean of her state party delegation in first session, minus the same measure in the second session.16 In short, if MOVEMENT TOWARD STATE is positive, the MC moved closer to her state party delegation between sessions; if it is negative, the MC moved further from her state party delegation; and if it is zero, the MC's relative distance from the state party delegation remained the same;

CONSTANT - is set equal to 1;

ASPIRANT - is the familiar dummy;

MAGNITUDE NOMINATE1 - is the absolute value of the MC's first session NOMINATE score; and

MAGNITUDE NOMINATE1*ASPIRANT - is an interactive term designed to pick up any difference in the effect of MAGNITUDE NOMINATE1 between aspirants and non-aspirants.

The results strongly support the hypothesis. One slight surprise is the negative coefficient on the ASPIRANT variable, indicating that aspirants near the center of the ideological spectrum tend to move slightlyfurther from their state delegations between sessions - even more than do non-aspirants. But the slope



13

coefficients for the MAGNITUDE variabled tell the expected story. Even among non-aspirants, the more extreme tend to move toward their state delegation between sessions. But among aspirants, this effect is more than three times as great. And the t-score for MAGNITUDE NOMINATE1*ASPIRANT shows that the effect on aspirants versus non-aspirants is statistically discernible at conventional significance levels. In short, the more extreme the aspirant, the more she tends to move toward the rest of her state delegation between the two sessions of her last House term.

Of course, access to statewide office is controlled first by the party, in providing the nomination, and then by the state electorate as a whole. The same logic that suggests that aspirants should move toward their state delegations suggests that they should move toward their state party delegations.17 When equation (3) is run to test for movement relative to state party delegations, rather than state delegations as a whole, the results are strikingly similar.18 Again, the more extreme the aspirant, the more she tends to move toward her state party delegation between sessions. And this effect is more than twice as great among aspirants than among non-aspirants.

The similarity of results in the tests for movement relative to state delegation and state party delegation are not surprising, largely because the state delegation by definition subsumes the party delegation. For this reason, the tests run so far cannot distinguish whether movement among aspirants represents an effort to court the state constituency for the general election, or the state party constituency for the primary. A test that simultaneously examines the effects of the relevant (state and state party) delegations on aspirant voting behavior is necessary to untangle these effects.

Equation (4), below, offers this simultaneous analysis. The OLS regression model used to test this hypothesis is an extension of equation (1), which determined the extent to which second session NOMINATE scores followed first session scores for aspirants versus non-aspirants. Here, that basic logic is ex- panded to test for the additional explanatory power of state party, and statewide, delegation scores. To the extent that aspirants to statewide office move closer to the state party delegation's mean NOMINATE score, they demonstrate an effort to court the state party to secure the nomination. To the extent that they move toward the mean NOMINATE score of the state delegation as a whole, they demonstrate an effort to court the state electorate. Move- ment on either front would account in a rational manner for the increased shirking against their districts that Zupan (1990) finds by aspiring House members. Movement on the first front would support Barro's (1973) speculation that parties control pensions for last term MCs - at least for those who plan to run subsequently for an office in their state.

The equation is as follows:



14

NOMINATE2 = - .001(CONSTANT) - .017(ASPIRANT) (-.615) (-1.40)

+ .978(NOMINATE1) (209) -. 108(NOMINATE1 *ASPIRANT) (-4.15) + .041(STATE DELEGATION AVG2) (6.27) -.010(STATE DELEGATION

AVG2*ASPIRANT) (- .270) -.007(PARTY DELEGATION AVG2) (-1.37) + .071(PARTY DELEGATION

AVG2*ASPIRANT) (2.32)

Adjusted R2 = .953

Here, the use of NOMINATE1 and NOMINATE2 as indicators of the extent to which second session voting follows first session voting is the same as in equation (1), as is the use of the term interacting NOMINATE1 with ASPIR- ANT. The terms STATE DELEGATION AVG2 and PARTY DELEGATION AVG2 are the purified means of the NOMINATE scores for the legislators from an MC's state, and state party, respectively. These terms measure the extent to which the ideological locations of an MC's respective delegations can explain variance in NOMINATE2 that is not explained by NOMINATE1. Fi- nally, the interactive terms corresponding to each delegation average pick up the extent to which the influence of state and state party delegations on NOMINATE2 is different for aspirants from that for non-aspirants.

The results tell a more complex story than do the previous regressions. The constant term and the ASPIRANT dummy demonstrate that second session scores are not significantly different from first session scores, and that aspirants are not significantly more liberal or conservative than are non-aspirants. As in Equation 1, the coefficients to NOMINATE 1 and to the interactive term NOMINATE 1 *ASPIRANT demonstrate clearly that NOMINATE2 follows NOMINATE1 less closely for aspirants than for non-aspirants. This reflects the degree to which aspirants alter their voting behavior more than do non- aspirants between sessions. The .041 coefficient to the STATE DELEGATION AVG2 variable indicates that among non-aspirants, the state delegation's voting in the second session exerts some pull on individual MCs; that is, some of the change in an MC's scores between sessions can be attributed to movement toward the state delegation mean. But the small value for the corresponding



15

interactive term is not significantly different from zero, suggesting that this pull is no stronger for aspirants than for non-aspirants. The effects for state party delegation, however, are exactly the reverse. The small and non- significant coefficient to PARTY DELEGATION AVG2 suggests that among non-aspirants, the voting pattern of the state party delegation does not exercise an independent influence on individual voting. But the relatively large (.071) and significant coefficient to the interactive term PARTY DELEGATION AVG2*ASPIRANT demonstrates that changes in voting behavior between sessions among aspirants to statewide office are systematically related to the positions of their state party delegations.

The results of this regression confirm and clarify those of the earlier tests. The story this evidence tells is one in which aspirants alter their voting behavior more during their last Congress than do non-aspirants; and more extreme aspirants alter their voting more than moderates. Equation 4 demonstrates that the relative position of an MC to her state delegation can explain a part of the variance in the member's voting between sessions, but that the effect of the state delegation per se is no greater for aspirants than for non-aspirants. Final- ly, it shows that the relative position of an MC's stateparty delegation explains a larger part of this variance for aspirants only.

6. Conclusion

This study contrasts two schools of thought on how last term shirking by MCs is controlled: the first focuses on electoral market forces, and the second on pensions. Those who make the strongest case for the efficiency of electoral markets argue that shirking among last term MCs against their constituencies is not significant. This would seem to imply that voting patterns among last term MCs should be stable and consistent with prior voting. The tests conducted here indicate, however, that for MCs who aspire to statewide office - particularly those who are ideologically extreme - this conclusion is not valid. Although I have not tested for whether changes in voting patterns among last term MCs represent systematic violations of district interests, it is clear that aspirants change their voting behavior more than do non-aspirants.

Those who are less confident of the efficiency of electoral markets have suggested that political parties might control post-congressional career opportunities. If such a pension system is effective, then one should expect some last term representatives to show increased sensitivity to party interests. The tests conducted here demonstrate that among MCs who aspire to statewide office, this is the case. And the more extreme the aspirant, the more dramatic the movement toward her state and state party delegations. This evidence supports the Madisonian hypothesis that larger electoral units should be more politically



16

moderate. And it supports in a straightforward manner the proposition that ambitious politicians are responsible to future principals, perhaps even at the expense of current principals.

If increased shirking against districts is accompanied by responsiveness to state parties, might we conclude that party-administered pensions can mitigate the last term problem? Is the last term problem one of agency shift, rather than agency loss? Although the evidence here demonstrates that such an agency shift exists, it falls far short of suggesting pensions as a solution to the last term problem. Most importantly, this study only shows that a pension effect exists for those representatives who leave the House to run for statewide office. This includes an average of only 3.2% of MCs for each Congress studied. More- over, an average of only 8.2% of MCs left office voluntarily at the end of each Congress for any reason.19

The limited effectiveness of pensions also suggests some comment on the prospects for electoral control where restrictions on congressional terms are imposed. The public choice literature on electoral markets demonstrates that the sorting effect currently limits the absolute level of shirking by last term politicians. Last term representatives are consistently found to shirk less than do cross-sections of their peers. One could derive from this the conclusion that term limits pose no threat to the control of politicians. Yet last term representatives may still shirk more than they did prior to their last term. The evidence presented here suggests strongly that this is the case for aspirants, because they alter their voting patterns more than do non-aspirants. Moreover, aspirants move toward their state parties between sessions of their last Congress. The evidence here, then, suggests that sorting alone does not ensure responsiveness of all last term politicians to their constituents' interests.

Although sorting is not perfect now, term limits would likely further under- mine its effectiveness. First, term limits would curtail the period over which sorting could work. Voters would have fewer opportunities to identify and defeat shirkers while rewarding naturally good agents with reelection. Second, term limits would impose an end-point to the game between representative and constituents that would be recognized by all parties before the game even be- gins. Thus, the pattern of mutual defection - of preemptive shirking and of throwing the bum out - would be encouraged.

If all of this could be mitigated by a party-administered pension system, then term limits would not necessarily imply a loss of electoral control. Individual MCs would be responsive to parties, which in turn remain responsive to the electorate regardless of term limits. If party pensions were perfectly effective, then the existence of term limits would imply that parties should be responsive to voter interests across the area for which parties administer pensions. Thus, if state parties administered perfect pensions, then responsiveness of MCs should exist at the state party - rather than district - level.



17

Evidence of the effectiveness of pensions, then, is evidence of the prospects for the electoral control of politicians under term limits. The evidence presented here suggets that pensions currently exist, but are effective under conditions far too limited to serve as a source of electoral control across all representatives.

Notes

1. Kau and Rubin (1979), Peltzman (1984), Ferejohn (1986), Lott (1987), Davis and Porter

(1989), Dougan and Munger (1989), Higgs (1989), Lott and Reed (1989), Lott (1990a and

1990b), VanBeek (1990) McArthur (1990), Kalt and Zupan (1990), Zupan (1990), Bender (1991), and Lott and Davis (1992) are the ones that will be considered here.

2. Public choice studies of shirking, including this one, embrace a delegate model of representation, as opposed to a trustee model (see Pitkin, 1967, especially Chs. 6 and 9). MCs are as- sumed to be shirking when they fail to act on behalf of the interests of their constituents, as their constituents perceive them. Representatives are not supposed to act according to some alternative conception of constituent interests based on superior knowledge or long-term strategy.

3. Dougan and Munger (1989) also use the relative voting records of senators from the same state to detect shirking.

4. The logic here is a variant of that of the Chain Store Paradox, described by Ordeshook (1986: 451-462). See also Barro (1973), Becker and Stigler (1974), and Kreps and Wilson (1982).

5. Becker and Stigler (1974) developed this arguement to explain how law enforcement officers

might be discouraged from engaging in corruption. 6. Of course, MCs receive federal pensions as well. But in order to disqualify oneself for these

pension payments, an MC has to go some distance beyond merely shirking on her constituents

(unless said constituents are below the age of consent, or some equally heinous crime). Thus, I shall only concern myself here with party pensions.

7. See Davis and Porter (1989) and Kalt and Zupan (1990) for cross-sectional analyses that reach the opposite conclusion - that shirking increases as retirement approaches. This remains a

point of serious contention. Lott and Davis (1992) challenge Kalt and Zupan's (1990) metho-

dology on a number of counts. Zupan (1990), however, provides a more focused and compel- ling argument that voting behavior changes among congressmembers in their last term than does Kalt and Zupan.

8. VanBeek's (1990) results contradict Zupan's. The last word in this debate surely has not been written.

9. In this case, the obligation was to what Bender (1991) characterizes as the "liberal agenda," rather than to indicators of district interests. The critical test Bender constructs to detect shirk-

ing, nevertheless, is impressive. 10. This argument is made at length by Schlesinger (1966). 11. The methodology used in compiling the data is discussed in detail in Appendix A. The statisti-

cal qualities of NOMINATE scores are discussed at greater length in Appendix B. 12. The relevant OLS regression is as follows:

DIFFERENCE = .096(CONSTANT) +.023(ASPIRANT) (70.5) (3.08) Adjusted R2 = .002

where



18

DIFFERENCE - is the absolute value of the difference between an MC's NOMINATE scores for the two sessions of the Congress:

CONSTANT - is set equal to 1; and

ASPIRANT - is a dummy variable scored 1 if an MC leaves the House after that term to run for a statewide office, and 0 otherwise.

13. The coefficient for aspirants is determined by simply adding the negative coefficient for the interactive term to the coefficient for NOMINATE1.

14. The total effect for aspirants is the value of the coefficient for non-aspirants (.044) plus the value of the coefficient for the interactive term, which indicates only the additional effect for aspirants, (.118). The total effect for aspirants, then, is .162.

15. If we run equation (2) separately for aspirant and non-aspirants, the results are:

Aspirants DIFFERENCE = .032 + .162(MAGNITUDE NOMINATE1); (2a) Non-Aspirants DIFFERENCE = .074 + .044(MAGNITUDE NOMINATE1); (2b)

The expected DIFFERENCE is equal for aspirants and non-aspirants alike when MAGNI- TUDE NOMINATE1 = .36. Below that score, aspirants' expected difference is less than that for non-aspirants. Above .36, aspirant's expected difference is substantially greater.

16. For an explanation of "purified means," and a general discussion of the voting and career data on MCs used here, see Appendix A.

17. It should not matter whether control over the nomination is administered by the party as an organization, or, more likely, by the rank-and-file party members through primary voting. Either scenario implies that aspirants to statewide office must court the general interests to which the entire state party delegation is responsive.

18. MOVEMENT TOWARD PARTY = -.095(CONSTANT) (-8.31)

- .049(ASPIRANT) (-0.78)

+ .14(MAGNITUDE NOMINATE1) (7.10)

+ .18(MAGNITUDE NOMINATE1*ASPIRANT) (1.90)

Adjusted R2 = .015

19. Given the extraordinarily high rate at which MCs are leaving their House seats - and in many cases running for statewide office - in 1992, the 102nd Congress may prove to be a high water mark in the importance of pensions in influencing voting behavior. Unfortunately for this study, the voting records of this group are not yet complete, much less available for analysis.

References

Barro, R.J. (1973). The control of politicians: An economic model. Public Choice 14 (Spring): 19-42.

Becker, G.S. and Stigler, G.J. (1974). Law enforcement, malfeasance, and compensation of enforcers. The Journal of Legal Studies 3(1) (January): 1-18.



19

Bender, B. (1991). The influence of ideology on congressional voting. Economic Inquiry 29: 416-428.

Davis, M.L. and Porter, P.K. (1989). A test for pure or apparent ideology in congressional voting. Public Choice 60(2): 101-111.

Dougan, W.R. and Munger, M.C. (1989). The rationality of ideology. Journal of Law and Eco- nomics 32: 119-142.

Fenno, R.F. Jr. (1978). Home style. Boston: Little Brown. Ferejohn, J. (1986). Incumbent performance and electoral control. Public Choice 50: 5-25. Higgs, R. (1989). Do legislators' votes reflect constituency preference? A simple way to evaluate

the Senate. Public Choice 63(2) (November): 175-181. Inter-University Consortium for Political and Social Research (Fall 1987). Roster of U.S. congres-

sional officeholders and biographical characteristics of members of the U.S. Congress, 1789-1987. (ICPSR 7803).

Kalt, J.P. and Zupan, M.A. (1990). The apparent ideological behavior of legislators: Testing for principal-agent slack in political institutions. Journal of Law and Economics 33(1) (April): 103-131.

Kau, J.B. and Rubin, P.H. (1979). Self-interest, ideology, and logrolling in congressional voting. Journal of Law and Economics 22: 365-384.

Kiewiet, D.R. and McCubbins, M.D. (1991). The logic of delegation: Congressional parties and the appropriations process. Chicago: University of Chicago Press.

Kreps, D.M. and Wilson, R. (1982). Sequential equilibria. Econometrica 50: 863-894. Lott, J.R., Jr. (1987). Political cheating. Public Choice 52(2): 169-186. Lott, J.R., Jr. (1990a). Attendance rates, political shirking, and the effect of post-elective office

employment. Economic Inquiry 28 (January): 133-150. Lott, J.R., Jr. (1990b). Why opportunistic behavior in political markets tends to be self-correcting:

An explanation for why senators from the same state vote so differently. Business Economics Working Paper #90-15 (July).

Lott, J.R., Jr. and Reed, W.R. (1989). Shirking and sorting in a political market with finite-lived politicians. Public Choice 61(1) (April): 75-96.

Lott, J.R., Jr. and Davis, M.L. (1992). A critical review and extension of the political shirking literature. Public Choice 74(4): 461-484.

Madison, James (1961). Federalist #10. In A. Hamilton, J. Madison, and J. Jay, The Federalist. New York: Signet.

McArthur, J. (1990). Congressional attendance and political shirking in lame duck sessions. Un- published draft (May).

Ordeshook, P.C. (1986). Game theory and political theory: An introduction. New York: Cam- bridge University Press.

Peltzman, S. (1984). Constituent interest and congressional voting. Journal of Law and Econom- ics 27: 181-210.

Pitkin, H.F. (1967). The concept of representation. Berkeley: University of California Press. Poole, K.T. and Rosenthal, H. (1985). A spatial model for legislative roll call analysis. American

Journal of Political Science 29(2): 357-384. Poole, K.T. and Rosenthal, H. (1988). Patterns of congressional voting. GSIA Working Paper

#88-89-07. Schlesinger, J.A. (1966). Ambition and politics: Political careers in the United States. Chicago:

Rand McNally. VanBeek, J.R. (1990). Does the decision to retire increase the amount of political shirking? Un-

published Draft (18 August). Zupan, M.A. (1990). The last period problem in politics: Do congressional representatives not sub-

ject to a reelection constraint alter their voting behavior? Public Choice 65(2) (May)167-180.



20

Appendix A: The data

I combined biographical profile data provided by the Inter-University Consortium for Political and Social Research (ICPSR) with NOMINATE scores (see Appendix B) for members, matching the data by ICPSR identification numbers. NOMINATE scores broken down by congressional sessions were available for the 89th through 98th Congresses, so my analysis is limited to this time span.

Coding

Identifying aspirants to state office proved somewhat troublesome using the ICPSR biographical data, because the best variable to distinguish whether a given member aspired to another office during a given Congress does not establish unequivocally whether she actually ran for a statewide office. "Why Member Left Congress Of Record" is coded such that a value of 5 means "Sought or accepted other office," and a value of 7 means "Went to Senate." (Codebook for ICPSR dataset 7803). A value of 5, then, is not exclusive to aspirants to statewide office; and a value of 7 is not exhaustive. Those scored at 5 could include aspirants to the presidency or other offices. On the other hand, those who ran for the Senate and lost, and those who ran for governor, would not be scored at 7.

Nevertheless, the best solution is to include as an aspirant any MC who scored a 5 or 7 on this variable, for the following reasons. Those who were appointed to any federal office would be scored a 6, and so would not be included among those who "accepted other office." Thus, those who could be scored at 5, but who did not aspire to statewide elected office, would include those members who left Congress to run for the presidency, or those who accepted appointments below the federal level, or those who actually left congress to run for an office with less than a statewide constituency. I am confident that those fitting into this last category are few in number during the period being studied, if any exist at all. In addition, for an MC to be appointed at the state level, it seems straightforward that her performance in the House would have to be acceptable to politicians in her home state. Since we are testing here for whether aspirants curry favor in their home state, and with their state party, at the expense of their district, such appointees would seem to fit the bill of aspirant well.

The coding ambiguity in the ICPSR data, then, seems problematic only for cases of presidential aspirants. However, it is important to keep in mind that only those who left the House to campaign for higher office are classified as aspirants. MCs who consider running for the presidency, and may even contest primaries, need not give up their House seat. Indeed only those who gain their party's nomination for another office need leave Congress. But over the period studied, no incumbent House member gained the presidential nomination of either party. Thus, by including as aspirants those who scored either a 5 or a 7 on "Why Member Left Congress Of Record," we are including all those aspirants to statewide office who succeeded in gaining nominations, and no MCs who do not face the incentives to court state and state party delegations.

Delegation averages

Mean NOMINATE scores for state delegations, and state party delegations, are used as indicators of state and party interests. The regressions test whether mean delegation scores act as better predictors of voting patterns among aspirants than among non-aspirants. Two adjustments had



21

to be made to the data, however, before mean delegation scores could be used. First, delegations with only one member were eliminated. Thus, in the regressions that test for movement relative to state delegations, those states with just one congressperson are removed. This left me with 4247 observations (MCs) over the ten Congresses I studied, including 137 aspirants. Likewise, in the regressions that test for movement relative to state party delegations, any state party delegation with only one MC is not included. This left me with 4108 observations over the ten Congresses, with 127 aspirants.

Second, the impact of the individual MC being tested had to be "purified" from the mean for each delegation, to eliminate autoregressive effects. If a given MC's NOMINATE score is included in calculating a mean for her delegation, and then that mean is subsequently used as a predictor of the MC's NOMINATE score, one should expect the measured impact of the mean to be biased upwards. To eliminate this effect, the NOMINATE scores of each MC were eliminated from the calculation of the mean scores for their corresponding delegations. The "purified" delegation means were then used as predictors of second session NOMINATE scores.

Appendix B: NOMINATE scores

NOMINATE scores are estimates of MC locations on a unidimensional, liberal-conservative scale. They are derived from each MC's votes on all non-unanimous roll calls taken in each session of Congress, by the Dynamic Nominal Three-step Estimation procedure described by Poole and Rosenthal (1985, 1988). Non-unanimous roll calls, for the purposes of NOMINATE, are all those in which at least 2.5% of MCs voted against the majority. The scores used here are scaled on a normal distribution, centered at 0, with standard deviation of .59. The minimum score is - 2.23 (the most liberal), and the maximum is 1.92 (most conservative). The vast majority of scores, however, fall between - 1 < NOMINATE < 1.

As measures of ideology, NOMINATE scores are superior to interest group rating scores used in other analyses for a number of reasons. First, NOMINATE scores are derived from all contested roll calls, not from smaller - and potentially biased - selections of votes chosen by interest groups. In addition, because interest group ratings are constructed to identify ideological friends and ene- mies, they often display an exagerrated, skewed distribution of MCs across the scale. (For a discussion of the relative strengths of NOMINATE scores vs. interest group ratings, see Kiewiet and McCubbins, 1991, Ch. 3.)

A related point is that in calculating NOMINATE scores, absences are treated as missing data, rather than being given some value determined by political interests implicit in the estimator. In calculating ADA scores, for example, absences on the relatively small number of key votes included in the estimation are regarded as conservative votes. Thus, estimates are distorted for those MCs with unusually high levels of absenteeism. MCs in their last House term - such as aspirants to statewide office - are just such a group (Lott, 1990a). Using NOMINATE, on the other hand, eliminates this problem because NOMINATE regards non-voting as non-information. And because NOMINATE scores are derived from a far larger, and non-biased, group of roll calls, they can be estimated with good precision even for MCs with higher than average rates of absenteeism.

In the end, NOMINATE scores have correlations above .9 with ADA scores, and with the coor- dinates derived from a set of interest group ratings compiled by Poole and Rosenthal (1985: 360). So clearly, using NOMINATE scores does not imply the use of a radically different metric of ideology from studies that use interest group scores. But NOMINATE does offer some distinct advan- tages in terms of comleteness and precision.

Another feature of the NOMINATE scores used here is that they are constrained at the extremes according to a function determined by the position of each legislator relative to all those who over-



22

lap with his service period. Some constraint is necessary, given the problems in estimation for those MCs who are the most ideologically pure, or extreme (Poole and Rosenthal, 1985). But the constraining ellipse employed by NOMINATE does not invite end point problems in estimation by truncating the scores at arbitrary boundaries. Rather, scores for extreme MCs are constrained by the relative positions of their contemporaries. Ultimately, the constraining mechanism has to be employed for only 4.2% of all legislators (Poole and Rosenthal, 1988).

NOMINATE scores have the additional feature of representing actual scalings of ideological positions rather than only relative scalings. This means that NOMINATE scores calculated jointly across more than one session of Congress can detect and measure aggregate shifts in ideology across all members, and that any such shifts can be distinguished from the effects of ideological shifts in legislative agenda. Because I am interested in movement of aspiring MCs relative to their state, and state property, delegations, I do not want to pic up statistical noise from whatever aggregate ideological shifts may take place among groups of MCs, such as delegations.

There are two straightforward ways of doing this. One would be to convert NOMINATE scores to percentile rankings that would reflect the relative ideological positions of legislators. This is done, however, only at the expense of forcing the normal distribution of scores into a uniform distribution of rankings. The uniform distribution necessarily sacrifices information about the relative ideological distances among MCs. In fact, when I run my regressions on percentile rankings, the results are consistent with those derived using raw scores, but the statistical significance of the coefficients is weaker, as would be expected. The other solution is to use NOMINATE scores calculated session-by-session, rather than jointly across sessions. Thus, for each session all individual scores, as well as the average scores of state and state party delegations, are in a metric unique to that session. Such a method ignores the issue of aggregate ideological shifts across congresses, but is particulary sensitive to the positions of MCs relative to their delegations. NOMINATE scores used here are calculated session-by-session.

The fact that such scores are each scaled in unique metrics, of course, poses a problem when scores from one session are used to predict scores in subsequent sessions. To the extent that there are shifts in ideology between sessions, we might expect predictive powers to be limited. In fact, however, Poole and Rosenthal find such shifts to be minimal, at both the individual and aggregate levels. Based on jointly calculated scores, individual MCs in the modern era tend to vary at 1-2% across congresses. And session-by-session scores correlate with jointly calculated scores at better than .98 (personal communication with Poole). So we should expect NOMINATE scores calculated by session to predict subsequent session scores with slightly less than .98 accuracy. This is precisely what the regressions run here find, across all MCs. Predictions based on prior session scores for MCs aspiring to statewide office, however, are significantly less accurate. The magnitude of the difference, and it uniqueness to aspirants, preclude the possibility that the existence of unique metrics could be causing the predictive inaccuracy. The inaccuracy can be explained, however, as significant shifts in voting patterns among aspirants.



Documents

Political Shirking and the Last Term Problem: Evidence for a Party-Administered Pension System